[Podcast] Intern S1: AI for Science скачать в хорошем качестве

[Podcast] Intern S1: AI for Science

Intern-S1: A Scientific Multimodal Foundation Model https://arxiv.org/pdf/2508.15763 Intern-S1 is a specialized open-source multimodal foundation model designed to bridge the gap between general artificial intelligence and professional scientific research. Developed by the Shanghai AI Laboratory, this 281-billion-parameter model uses a Mixture-of-Experts (MoE) architecture and was trained on over 2.5 trillion scientific tokens to master complex data like molecular structures and time-series signals. Innovative features such as a dynamic tokenizer and a Mixture-of-Rewards (MoR) reinforcement learning framework allow the system to process domain-specific information more efficiently than general-purpose models. Evaluations demonstrate that it outperforms both open-source and many closed-source counterparts in tasks like chemical reaction prediction and thermodynamic stability analysis. By releasing the model weights and training toolkits, the developers aim to accelerate global breakthroughs in high-value scientific fields. #ai #research #largelanguagemodels